I think each piece does it's job well, and better than if someone tried to somehow consolidate multiple web languages into one.
HTTP has proven itself robust and versatile, especially with the wide spread adoption of REST development strategies. It also allows you free choice of language to implement on the server. Before the popularity of web development took off, language choice was largely circumscribed by the environment to which you wanted to deploy (the current controversy over iPhone OS hearkens back to those days).
That's true of JavaScript in the browser today, but Javascript has proven flexible enough to allow for many different ways of programming in the browser (Cappuccino, jQuery, etc.). And simply by the size of its installed base, it is inspiring innovations in dynamic language performance that remind me of what happened with Java and byte-code virtual machines.
HTML and CSS split the job of content form and appearance nicely, and the speed of HTML5 adoption by competing browsers has been impressive, in my opinion. I think that the CSS approach of separating styling from the content is superior in many ways to traditional desktop development.
If I were designing a web application platform on purpose, major design criteria might be: 1) data and apps hosted remotely, 2) discoverable apps/services, 3) hypertext-style linking of resources, 4) ability to fully exploit client CPU and GPU power for UI, and for compute where appropriate, 5) consistency with common desktop UI concepts (like drag/drop), 6) flexibility in languages and developer tools, allowing for compiled, statically checked languages for those who like them. With the current web, I think we have 1-3, but not 4-6. (I think I would trade the enforced niceties of HTML/CSS for more flexibility in the UI.)
But... isn't that called a native Java app?
XAML and C# are rather close in comparison. The capabilities of HTML and JS are only limited by what the browser allows.
I understand your point, but I don't think the HTML/JS/Server Side language setup is far off the mark.